This study develops tsunami evacuation plans in Padang, Indonesia, using
a stochastic tsunami simulation method. The stochastic results are based on
multiple earthquake scenarios for different magnitudes (
Tsunami hazard and risk assessments have become an important issue in
tsunami-prone regions especially after the 2004 Aceh–Andaman earthquake
(
Padang is a home of more than 850 000 and is one of the most urbanized
cities in western Sumatra. The 2004 Aceh–Andaman tsunami did not
significantly affect this region since the earthquake source was
Moreover, an effective tsunami evacuation plan should combine both horizontal evacuation to high grounds and vertical evacuation to designated tsunami-resistant shelters (FEMA P-646, 2012). In the coastal areas where people can afford a relatively short evacuation time (less than 30 min), the vertical evacuation is highly desirable (Scheer et al., 2012; Wood et al., 2014). In previous investigations, the tsunami arrival time in Padang was estimated to be 20–30 min (Borrero et al., 2006; McCloskey et al., 2008; Schlurmann et al., 2010; Muhammad et al., 2016) and, therefore, effective vertical evacuation plans are needed in this region. In Padang, 23 tsunami evacuation shelters (TESs) have been planned and built in the urban areas near the coastal line. However, an extensive assessment of the TESs in Padang during the tsunami event was not conducted by the previous studies (Schlurmann et al., 2010; Muhari et al., 2010, 2011; Imamura et al., 2011). Only horizontal evacuation time maps to safe areas were provided in Schlurmann et al. (2010), excluding seismic and tsunami vulnerability assessments of TESs. Subsequently, it is important to assess the TESs considering impacts of ground shaking and tsunami in Padang as part of future tsunami mitigation planning. The results of the TES assessment can be further used to develop integrated horizontal–vertical tsunami evacuation maps.
Procedures of stochastic tsunami simulation.
Building upon the previous studies, this study develops tsunami evacuation
plans in Padang based on the stochastic tsunami simulation method. This
approach generates multiple earthquake scenarios by considering the
uncertainties of the earthquake source parameters and slip distributions,
both of which have major influence on the tsunami hazards. Hence, it is
suitable to estimate the tsunami hazard level in Padang. Regional
seismological characteristics are taken into account in generating stochastic
earthquake scenarios based on the finite-fault models of the past earthquakes
in the Sunda subduction zone. In addition, a 5
The stochastic tsunami simulation method (Muhammad et al., 2016) is adopted to estimate the tsunami hazard level in Padang. To generate earthquake source model stochastically, earthquake scenarios in terms of magnitude and source region need to be set up in advance. An appropriate model of fault rupture zone including geometry of the fault plane and asperity regions is then defined (see Fig. 2). The geometry is essential to outlining the earthquake source zone, whilst a so-called asperity zone determines the areas of concentrated slips within the fault plane. In general, modeling an earthquake rupture process in terms of earthquake source and asperity zones for the future tsunamigenic earthquakes in the Mentawai–Sunda region is complicated and has significant uncertainty (Natawidjaja, 2006; Griffin et al., 2016). In this study, the future earthquake source area in the Mentawai segment is defined using the fault rupture areas of the historical subduction earthquakes in the Sunda subduction zone (see Muhammad et al., 2016, for details).
Firstly, a generic fault model covering the entire region of the Mentawai
segment with a length of 920
Secondly, within the fault plane of the source zone, asperity zones are set
up. The asperity zones reflect the regional seismological knowledge of
earthquake ruptures. Understanding the rupture process of past seismicity in
the Mentawai segment is essential to determine the asperity zones. Based on
the past seismicity, the most likely asperity zones for the future
tsunamigenic earthquake events in the Mentawai segment are located in the
rupture areas of the 1797 and 1833 events since these two events were the
last tsunamigenic earthquake events in the Mentawai–Sunda region (Borrerro
et al., 2006; Natawidjaja et al., 2006; McCloskey et al., 2008). A 300
The stochastic tsunami simulation involves two main processes: (1) stochastic
earthquake source model generation and (2) Monte Carlo tsunami simulation
(see Fig. 2). For generating realistic source models stochastically, regional
seismological characteristics of Sumatra are analyzed. It is carried out by
estimating earthquake source properties including geometry (fault length,
Subsequently, the calculated regional earthquake source parameters are compared against the global scaling relationships developed by Goda et al. (2016) to examine the adequacy of the global models to the Mentawai–Sunda region. Muhammad et al. (2016) concluded that the regional earthquake source parameters calculated from the 19 past Sunda earthquakes are in good agreement with these scaling relationships; subsequently, the global models are adopted in this study. A set of 100 source models is then generated for each magnitude. Therefore, the total number of the stochastic earthquake slip models used in this study is 300.
For a given stochastic source model, the initial deformation of the seabed is
calculated by considering both horizontal and vertical displacements of the
seafloor using Okada (1985) and Tanioka and Satake (1996). Tsunami modeling
is then carried out by solving nonlinear shallow water equations with run-up
(Eqs. 1 to 3). A finite-difference method implementing a staggered
leap-frog scheme is adopted to solve the governing equations (Goto et al.,
1997). In addition, in Goto et al. (1997) the moving boundary approach
developed by Iwasaki and Mano (1979) is used for inundation modeling. This
method has been successfully used to run the tsunami simulation in several
regions, including Padang, Indonesia, Mexico, and Japan (Muhari et al., 2010,
2011; Goda et al., 2014; Mori et al., 2017).
Digital elevation model for Padang:
An accurate DEM is essential for tsunami simulation and is particularly
critical in calculating tsunami inundation depths. Recently, several
international organizations and consortia have produced global DEM datasets,
including GTOPO30, SRTM30, SRTM3v2, SRTM3v4, SRTM1, and GDEM2. Currently, the
SRTM1 (
In this section, two global DEM datasets, i.e., SRTM1 and GDEM2, are adopted
to study the effect of DEMs on tsunami inundation modeling in the Padang
areas (see Fig. 4b and c). The elevation differences between the DEM5 and the
two global datasets (i.e., SRTM1 and GDEM2) are firstly assessed and
discussed. Then, the differences on tsunami inundation results from these
three datasets are further presented. Note that the bathymetry dataset
(GEBCO2014) is common for all three cases. The considered source model is one
realization from the 100 stochastic earthquake sources in the Mentawai–Sunda
zone for the
Statistics of elevation differences between global DEM datasets (i.e., SRTM1 and GDEM2) and the reference data (DEM5).
The elevation differences of the SRTM1 and the GDEM2 datasets with respect to
the reference DEM5 data are presented in Fig. 5a and b, respectively. The
elevation differences are calculated by taking the elevation differences
between the global DEM datasets (i.e., SRTM1 and GDEM2) and the reference
data. Several statistics, including the minimum, maximum, mean difference,
absolute mean difference, and root mean square error (RMSE) values, are
calculated for each global DEM dataset and presented in Table 1. The
difference with the reference data in terms of maximum elevation value for
SRTM1 is 7
Elevation differences of global DEM datasets with respect to DEM5:
Inundation areas in Padang for
Tsunami evacuation shelters in Padang.
Subsequently, to evaluate the effect of DEMs on inundation modeling, Fig. 7
illustrates inundation maps in Padang based on three different DEMs, i.e.,
DEM5 (Fig. 7a), SRTM1 (Fig. 7b), and GDEM2 (Fig. 7c). Total inundation areas
are also presented in Fig. 7. The inundation maps indicate that the global
DEM datasets (i.e., SRTM1 and GDEM2) underestimate the inundation areas
significantly. The total inundation area of SRTM1 is less than a half of DEM5
(i.e., 7.32
Prior to developing tsunami evacuation plans, structural vulnerability of TESs in Padang needs to be evaluated by considering both ground shaking and tsunami hazards. Suitable seismic and tsunami fragility models are used to assess the vulnerability of the TESs by determining the probability that a building could sustain a certain damage state for a given earthquake scenario. Since the ground shaking affects the TESs before the tsunami, the seismic vulnerability assessment of TESs is carried out first, followed by the tsunami vulnerability assessment. In this study, the combined effects of earthquake shaking and tsunami are not taken into account, because such multi-hazard fragility models are not available for TESs in Padang. Detailed procedures for the earthquake–tsunami vulnerability assessments of TESs are presented in the following sections.
Seismic intensity measures for a given earthquake scenario at a TES can be
effectively estimated using ground motion prediction equations (e.g.,
The formula to model ground motion intensities due to interface subduction
earthquakes is given by
To simulate ground motion intensities using the Abrahamson et al. model,
three parameters are needed as inputs: magnitude, rupture distance, and
average shear wave velocity for the considered site. For the TES assessment,
only the worst magnitude earthquake scenario is considered
(
The seismic vulnerability of TESs can be assessed using seismic fragility
models. The fragility models relate the probability that a building's damage
state exceeds a particular threshold for a ground motion parameter, such as
PGA and Sa (Rossetto and Elnashai, 2003). In general, the seismic damage
states (DS) can be categorized as slight (DS1), moderate (DS2), extensive
(DS3), and complete (DS4) (see Table 3). Typically, for DS1, fine cracks in
plaster partitions and infills to hairline cracking in beams and columns near
joints ( The TESs are designed and constructed according to the new Indonesian Earthquake Resistance Building Code (SNI-1726: 2002 and 2012), which adopted the US seismic design documents, i.e., FEMA P750
(2009),
regarding seismic design provisions for new building and other structures,
and ASCE/SEI 7-05 for the minimum design load criterion (ASCE, 2006; SNI-1726: 2012;
Kurniawan et al., 2014; Wijayanti et al., 2015; Aulia, 2016; Sengara et al.,
2016). HAZUS is a well-established earthquake loss estimation framework and has been implemented in several earthquake-prone countries for
seismic risk assessment purposes, e.g., Haiti, Puerto Rico, France, Romania,
Austria, and Indonesia (Kulmesh, 2010; Peterson and Small, 2012; Wijayanti
et al., 2015; Sengara et al., 2016).
In HAZUS, building performance under seismic actions is evaluated based on
the CSM. The CSM compares the structural capacity in terms of capacity curve
with the seismic demand on a structure that account for post-yielding
inelastic behavior of the structural system. The nonlinear inelastic
behavior of the structural system is taken into account by applying the
effective reduction factor to the elastic response spectrum (ERS) of the considered
earthquake scenario (Freeman, 2004; Kim et al., 2005; Monteiro et al., 2014).
The performance point, which is the expected seismic response of a structural
system for a given earthquake scenario, can be obtained graphically as an
intersection point of the capacity and demand curves. Figure 6a illustrates
the procedure for developing an inelastic response (demand) spectrum from the
elastic response (input) spectrum in HAZUS. First, the acceleration response
spectrum is generated from the earthquake simulation (see Sect. 2.2.1) and
is further converted into the acceleration–displacement response spectrum
(ADRS). In the CSM, the ADRS is defined as the ERS. Second, the inelastic demand spectrum is calculated by dividing the
ERS by the reduction factors (i.e.,
Finally, seismic fragility curves implemented in HAZUS are used to define the
damage functions of the building; typically, the fragility functions are
defined using the lognormal distribution. Four damage states – slight,
moderate, extensive, and complete – are defined in HAZUS (see Table 3 and
Fig. 6b for the descriptions). Subsequently, to determine whether a TES can
be used for post-earthquake tsunami evacuation purposes (not for shelters),
the building is categorized into safe and unsafe by referring to existing
tagging criteria (FEMA 356, 2000; HAZUS, 2003; Bazzurro et al., 2006)
including the following (see Fig. 6b)
Green tag: the building may have experienced onset damage but is safe for immediate occupancy. The none-to-slight damage state is
applicable. Yellow tag: re-occupancy of the building is restricted and limited access only is allowed. Moderate-to-extensive damage state
corresponds to this case. Red tag: the building is unsafe and no access is granted and will be completely damaged or collapsed.
Based on the above tagging criteria, the tsunami evacuation building may be
judged as unsafe for evacuation if the probability of extensive and complete
damage states is over 50 %. This assumption gives a 50–50 chance that
the building may experience above or below extensive damage (Bazzurro et al.,
2006). Moreover, the 50 % probability of extensive or severe damage
is typically identified as the threshold value of a yellow tag in HAZUS
that is adopted in this study (see Fig. 6b) and, hence, may be regarded as the
limit state to define the accessibility of buildings for emergency evacuation
during the tsunami inundation.
Once the TES is judged to be accessible for tsunami evacuation through the seismic vulnerability assessment, a tsunami vulnerability assessment is then carried out. The tsunami fragility models evaluate the probability of experiencing certain damage states for a building due to tsunami. The tsunami damage criteria for buildings are defined by the MLITT (2013) and are shown in Table 4.
In current literature, most tsunami fragility models have been developed
using post-tsunami survey data, remote sensing data, and numerical modeling
(e.g., Dias et al., 2009; Koshimura et al., 2009; Suppasri et al., 2011). In
this study, the model by Suppasri et al. (2011) is adopted for the following
reasons. It was developed through extensive remote sensing and tsunami survey
data (i.e.,
Subsequently, when the seismic and tsunami vulnerability assessment confirmed
that the TES is operational for evacuation, the TES capacity for
accommodation is calculated. This is evaluated by calculating the TES
capacity (TESC):
Floors of the TES buildings are categorized into two floor types: unsafe floor and evacuation floor. The unsafe floor is the
floor that is inundated by the tsunami, whilst the evacuation floor is the
floor for evacuation areas. Moreover, the inundated floor is considered to be
unsafe for evacuation and thus is excluded from building capacity
estimation during the tsunamigenic event. Space needed per person (SpP) at the evacuation areas in each TES building is 1 Existing evacuation area in each floor of the TES building is assumed to be equal for all floors because only total evacuation area
data for the whole building are available.
Inundation depth maps in Padang:
Inundation areas above 1
Next, horizontal and vertical tsunami evacuation time maps are developed
based on the total evacuation time (TET), which is calculated by summing
initial reaction time (IRT) and evacuation time (ET).
Finally, the integrated evacuation time maps are developed by combining horizontal and vertical evacuation time maps. The integrated evacuation time maps are calculated by taking a minimum evacuation time between evacuations horizontally and vertically. These maps are essential for the rescue teams to consider both evacuation options and subsequently may reduce the casualties during the tsunami event.
The main results that are discussed in this section focus on the tsunami
hazard level in Padang produced from all earthquake scenarios
The tsunami hazard level in Padang is investigated by assessing the tsunami
height and depth produced from the stochastic tsunami simulations for three
magnitude scenarios. The height presented in this study is the height of
water flow above the mean sea level, whilst the depth refers to the water
flow height above the ground. Firstly, the maximum inundation depth maps for
all scenarios along with the maximum inundation depth maps for the areas
above 1
The maximum inundation depth maps in Padang are presented in Fig. 8, whilst
the total inundation areas for the depth above 1
Moreover, the tsunami effects are found to be much more significant for the
To capture the variability of inundation extent in Padang, Figs. 10–13 show
the inundation height profiles along the coastal line and three rivers in
Padang. The length of the Kuranji, Banda Bakali, and Arau rivers from their
river mouths are 1.4, 1.7, and 2.45
Description of seismic damage state (HAZUS, 2003; Rosetto and Elnashai, 2003).
Damage state definitions for tsunami (Suppasri et al., 2014; Charvet et al., 2017).
Assessment of tsunami evacuation shelters (TES) in Padang in terms of depth, tsunami arrival time, and number of tsunami destructive events.
TES building capacity during the tsunami event considering the
Currently, in east and north of Padang, 23 TESs have been set up or designated by
the National Agency for Disaster Management (BNPB) of Padang (see Table 2).
However, their adequacy as emergency evacuation building was evaluated under
the design scenarios only. Consequently, re-assessment of TESs in Padang is
highly desirable by taking into account uncertainties associated with the
tsunami hazards. This will provide residents and emergency/rescue teams with
valuable information regarding the current tsunami risk exposure in Padang.
In this section, the tsunami inundation depth variability at each TES
building is firstly discussed. The vulnerability of each TES is then assessed and
discussed by considering the worst seismic and tsunami scenarios
(
Inundation depth variability at TES stations for
First, the variability of tsunami inundation depth at each TES is shown in
Fig. 14. The tsunami impacts to all TESs are insignificant in the case of
Second, the seismic vulnerability assessment results are carried out for the
The assessment that is illustrated in Fig. 16a–c ignores the inherent
uncertainty of input ground motions. To account for this uncertainty, ground
motion simulation results for 100 tsunamigenic earthquake scenarios are
presented in Fig. 16d and e by considering the prediction error terms of the
ground motion model together with inter-period correlations. The spectral
acceleration profiles show a range of ground shaking that is expected to
occur in Padang due to the 100 tsunamigenic earthquakes generated from the
Mentawai segment of the Sunda subduction zone. The range of Sa in Padang is
between 0.2 and 1.1
Fourth, the tsunami vulnerability assessment is performed. Using the maximum
inundation depths at all 23 TESs from the 100 earthquake scenarios of the
Subsequently, the estimation of TES building capacity is evaluated. This may
capture another point of view regarding the adequacy of existing TESs for
evacuation. Table 6 presents the estimation of TES building capacity during
the tsunami event considering the 10th percentile, the 50th percentile, and
the 90th percentile. In terms of capacity, except for shelters 16 and 17, all TES buildings can be used for vertical evacuation
during the 10th rank event. However, for the 50th percentile case, the
shelter number 1 (sport center of UNP) may not be operational, whilst for the
90th percentile case shelters 1 and 15 (Elementary school of 24
Padang) are unable to accommodate evacuees since all floors will be
inundated. Note that, for shelter number 1, there is only one floor since
most of the building areas are used for the sport arena. In terms of
capacity, for the 50th and 90th rank cases, the possible maximum capacity to
be accommodated at all TES buildings is only about 64 000 and 41 000
people, respectively. These numbers are insufficient in comparison to the
total population in the coastal region of Padang (i.e.,
This section presents the tsunami evacuation maps based on the stochastic
tsunami inundation depths in Padang. The developed tsunami evacuation maps
consist of tsunami inundation maps and horizontal, vertical, and integrated
evacuation time maps to the safe zones. Three scenarios are considered to
develop tsunami evacuation time maps for the
Figure 8c shows tsunami inundation maps in Padang, whilst Fig. 17 illustrates
tsunami evacuation time maps to the safe zones in Padang for the 10th
percentile (Fig. 17a), the 50th percentile (Fig. 17b), and the 90th
percentile (Fig. 17c). The horizontal tsunami evacuation time to the safe
areas is calculated and is used to produce the tsunami evacuation time maps.
The evacuation speed during the disaster event is chosen as
0.91
Based on the TES assessment results, the vertical and integrated evacuation
time maps are further developed to investigate the possibility of reducing
evacuation time to the safe areas. Figure 17d and e show the vertical and
integrated tsunami evacuation time maps to the TES locations. The shelters 1, 15, 16, and 17 are excluded while vertical and integration tsunami
evacuation time maps are developed because those shelters may be unsafe and
inundated during the worst tsunamigenic event (the 90th percentile of the
The main purpose of this study was to develop effective tsunami evacuation
plans in Padang based on the stochastic earthquake scenarios. Rigorous
tsunami hazard assessments in Padang have been carried out using a novel
stochastic tsunami simulation method to estimate the tsunami hazard level in
Padang using three magnitude scenarios including
For the
Lastly, although assessments for developing evacuation plans in Padang have
been conducted in this study using the results of rigorous stochastic tsunami
simulations, some limitations need to be addressed in future studies. These
include (1) tsunami hazard simulations being conducted using
high-resolution DEM (e.g., 10
The bathymetry and elevation data for the Sumatra region were obtained from the GEBCO2014 database
(
The authors declare that they have no conflict of interest.
The first author is grateful to the Directorate General of Resources for Science, Technology and Higher Education, Ministry of Research, Technology and Higher Education of Indonesia, which sponsored his PhD study. This work is also funded by the Engineering and Physical Sciences Research Council (EP/M001067/1). The authors are grateful to Sigit Sutikno, who provided the TES survey data in Padang. Edited by: Ira Didenkulova Reviewed by: Rachid Omira and one anonymous referee